Accurate and Efficient Optical Fiber WDM Transmission Modeling Using the Encoder-Only Transformer with Feature Decoupling Distributed Method

As the field of optical communication approaches the threshold of spectrum efficiency, the optical fiber transmission system adopts an increased number of Wavelength-division multiplexing (WDM) channels to achieve higher system capacity. Nonetheless, the classical split-step Fourier method (SSFM) for simulations in wide-band scenarios incurs substantial time consumption, thus constraining its potential applications in system evaluation and design. To overcome this limitation, we propose an innovative approach utilizing an encoder-only Transformer model with the feature decoupling distributed scheme and training using multi-span datasets for optical fiber channel modeling in the WDM scenarios. Our proposed model demonstrates excellent scalability, particularly well-suited for long-haul WDM transmission. In addition, the model yields a notable 1.44 dB improvement in the modeling Q factor performance for 21-channel 1200 km transmissions, outperforming the conventional BiLSTM approach. Moreover, the implementation achieves high efficiency, with a running time merely equivalent to 0.3% of the SSFM. Our methodology showcases high accuracy and effectiveness in facilitating research on WDM long-haul transmission systems.